Table 1 presents characteristics of the study participants. The sample was fairly representative of the general German population in terms of age and gender. The corresponding percentages for the age groups (18-29 y., 30-39 y., 40-49 y., 50-59 y., 60-69 y. and ≥70 y.), taken from the census (Destatis, 2015), are as follows: males: 18.1%, 15.2%, 21.4%, 17.3%, 13.5%, and 14.6%, and females: 16.4%, 14.0%, 19.4%, 16.3%, 13.4%, and 20.4%, respectively.

Age and gender differences

Figure 1 and Table 2 present mean scores of males and females, stratified in age groups. The age categories in Table 2 were condensed in order to get higher sample sizes for each category. The present health states decreased with age, and males reported better health states than females. The two-way ANOVA results were as follows: age (F=334.3, p<.001), gender (F=13.1, p<.001), age x gender (F=0, p=.998). The regression of present health on age and gender yielded the following equation:

Figure 1.

Self-rated health, broken down by age group and gender.

(0.09MB).

Table 2.

Mean scores of health assessments by age group and gender.

		Total	Males				Females				d	d
			≤ 40	40-59	≥ 60	all	≤ 40	40-59	≥ 60	all	(sex)	(age a)
N		2,409	325	440	357	1,122	365	472	450	1,287
Own health	Mean	76.2	90.2	78.5	65.8	77.9	87.5	75.8	63.1	74.7	0.16	1.43
	SE(Mean)	0.42	0.73	0.90	1.09	0.61	0.77	0.88	0.93	0.58
	Median	80.0	95.0	84.0	70.0	80.0	90.0	80.0	67.0	80.0
	SD	20.6	13.2	18.8	20.6	20.4	14.8	19.2	19.8	20.7
Patient A	Mean	42.7	42.0	43.2	41.3	42.2	42.7	44.5	42.2	43.2	0.05	0.03
	SD	19.1	19.8	19.0	18.3	19.1	19.6	19.8	17.9	19.1
Patient B	Mean	44.9	45.9	45.0	45.2	45.3	44.6	45.0	44.1	44.5	0.05	0.03
	SD	17.6	17.8	17.3	17.1	17.4	18.4	18.4	16.7	17.8
Patient AB	Mean	43.9	44.0	44.2	43.4	43.9	43.7	44.8	43.3	44.0	0.01	0.04
	SD	14.8	15.5	15.0	13.7	14.8	15.6	15.0	14.3	14.9

Note. d: Effect size; a: Comparison between youngest and oldest age group; SE(Mean): Standard error of mean.

Present health=105 – .593 * age + 2.8 * gender.

Gender has to be coded with 0 for females and 1 for males. The multiple R of this analysis was .51. For example, a 55-years old man has an expected present health score of 105.0 – 32.6 + 2.8=75.2. All main effects and interaction effects of age and gender for the health assessments of the vignettes (Table 2) were not statistically significant (p > .05 in all cases).

The first aim of the study was to test whether there are positive or negative correlations between respondents’ self-assessed health and their assessments of vignettes’ health. There were small positive correlations, with a coefficient of r=.12 between self-assessed health and assessment of the combined vignettes. Though the coefficient is statistically significant, the explained variance (1.44%) is small. The first conclusion of this positive correlation is that the positional (response shift) effect is not stronger than the judgment effect. However, conclusions about the extent of both effects are not possible.

According to the hypothesized positional (shift) effect, we expected that people suffering from chronic diseases (compared to healthy people) would evaluate the vignettes as being healthier. This hypothesis was not supported; the differences in Table 3 were small in magnitude. The comparison between the two regression models also provides only weak evidence for the positional (shift) effect: The inclusion of the number of chronic diseases in the model resulted in a very small change of the beta coefficient from .12 to .08. The latter (positive) coefficient cannot be directly compared with that reported by Grol-Prokopczyk et al. (2011) because that paper was based on ordered Probit regressions, but in that study the coefficient was also positive.

When the analyses were restricted to a subsample of subjects being in a rather bad health situation in terms of chronic diseases (and, therefore, similar to the persons described in the vignettes), the associations between own health and vignettes assessments increased, from r=.12 to r=.29. Such an effect was also observed in the study examining life satisfaction (Angelini et al., 2014). It indicates that in a sample of people with health problems the judgment effect clearly prevails.

This judgment effect or “health optimism” effect is related to the acquiescence effect or yes-set effect, the general tendency of people to provide affirmative answers to items, regardless of the context of the items (Billiet & McClendon, 2000; Ferrando & Lorenzo-Seva, 2010). When all items of a questionnaire are formulated in the same direction (e. g., high scores represent good health), the acquiescence effect contributes to a positive correlation between the items. In our study, the assessments of the vignettes and the self-rated health were measured with the same scale and the same orientation, where high scores indicated good health. In this case, we cannot distinguish between the acquiescence effect and the health optimism effect, since both effects represent the tendency to choose high scores on the health scale. The acquiescence effect provides an explanation for the fact that obviously opposite variables, e. g., positive and negative affectivity (Watson & Clark, 1997) or optimism and pessimism (Herzberg, Glaesmer, & Hoyer, 2006), nevertheless show only small negative correlations. These effects should also been taken into account when the influence of value orientations and protective factors on self-reported health is studied (Maercker et al., 2015).

We found similar results for the assessments of both vignettes regarding mean scores and lack of substantial correlations to self-rated health, though both vignettes were designed to be different. While vignette A was mainly characterized by physical problems, vignette B presented mental problems. The mean values of the assessments (M=42) are in the expected range, but the correlation between the assessments (r=.31) was only moderate. That is, the respondents judged them differently; some respondents attributed a higher level of health to vignette A, while others judged the opposite way. Nevertheless, there was a positive correlation.

There are several options for designing the vignettes. The description can be more or less complex, and the vignettes can be labeled as patients or people. The vignettes’ age and gender can be fixed, left open, or the respondents have to assign their own age and gender to the vignettes. Furthermore, the description of the vignettes can be tailored to the specific patients under study, as was done in the study with prostate cancer patients (Korfage et al., 2007). Finally, the symptoms described in the vignettes can be more or less severe. We decided to use two complex vignettes with no specification of disease or age. The disadvantage is that it is not clear which aspects of health are evaluated by the respondents, and which age they assigned to the vignettes. In contrast to our results, the Grol-Prokopczyk et al. (2011) study found a stable gender effect on “health optimism”: females assessed the vignettes as being healthier compared with the judgments of males. One possible reason for this difference is that in the Grol-Prokopczyk et al. study the vignettes were designed as “peers”, having the same age and gender as the respondent. Therefore, women had to assess female vignettes, which might contribute to the higher health ratings in that study. A recent study (Au & Lorgelly, 2014) compared “restricted” vignettes, described in terms of only one dimension, e. g., physical functioning, (Type A) with “complex” vignettes, described in terms of the five dimensions of the quality of life questionnaire EQ-5D (Type B). The participants of the study (Au & Lorgelly, 2014) reported that their subjectively rated response consistency (using the same scale for own assessment and the assessment of the vignettes) was higher for the complex type B (49%) compared with the restricted type A (29%).

Future research on vignettes should systematically investigate the effects of the factors complexity, symptom severity, and self-relatedness (vignettes of unspecified or fixed age and gender vs. age and gender of the respondent). This would help design most appropriate vignettes for the assessment of judgment effects. In addition, responding to vignettes in samples of patients may be different from that in a sample of the general population, and this relationship should also be systematically studied.

There are several options for the assessment of health. We used a 0-100 scale for rating the health states of both the respondents and the vignettes. Another possibility, used in the international SHARE study, is a five-point Likert scale (cf. Hartley, 2014) with the options: excellent, very good, good, fair, and poor (Grol-Prokopczyk et al., 2011; Jurges, 2007). Other researchers prefer the five answer labels: very good, good, fair, sometimes good/poor, and poor (Galenkamp, Deeg, Braam, & Huisman, 2013). Moreover, in other studies, the respondents were asked to rate health problems with one of five options: none, mild, moderate, severe, and extreme (Guindon & Boyle, 2012; Salomon et al., 2004). The advantage of the 0-100 scale is that it can be considered a metric scale, while the other examples require nonparametric statistics (Perneger, Gayet-Ageron, Courvoisier, Agoritsas, & Cullati, 2013).

The mean scores of the health status, given in Table 2, can be used as reference values in evaluating the health status of certain groups of patients. A previous German investigation using the same scale (Hinz, Klaiberg, Brähler, & König, 2006) found a mean score of M=77.1 which is very similar to the mean score of this study (M=76.2). The linear age effect and the lack of interactions between age and gender supported a linear regression approach. The coefficients of the regression can be used to calculate the expected health level of any age and gender group. The regression coefficients indicate that males report better subjective health than females (diff=2.8 points), and each age decade has a health worsening effect of 5.9 points on the 0-100 scale. In the previous study (Hinz et al., 2006), the gender effect was 3.9 points, and the effect of one age decade was 4.6 points.

Some further limitations of this examination should be mentioned. We designed the vignettes with relatively complex descriptions and severe symptoms. Vignettes A and B were referred to as “patient A” and “patient B”; perhaps the word “patient” triggered another category. It might be useful to return to more simple descriptions of one aspect of quality of life, or to vignettes with less severe symptoms. In the assessment of chronic conditions, we did not obtain clinical diagnoses. Therefore, our results are restricted to relationships between self-assessment data; the number of chronic conditions as measured with the SCQ is only a rough proxy of objective health. The possible selection bias has already been mentioned, though the age and gender distribution of our sample was similar to that of the German adult population.

In summary, we found only very small correlations between own health assessment and the assessment of complex vignettes, indicating that the hypothesized frame-of-reference effect was not strong enough to be used as a correcting factor for self-assessed health scores, even after controlling for a proxy of objective health.